실내 환경 내 반복 패턴 극복을 위한 텍스트 활용 시각적 장소 인식 방법 = Visual Place Recognition by Integrating Textual Information to Overcome Repetitive Patterns in Indoor Environments

다국어 초록 (Multilingual Abstract)

.

다국어 초록 (Multilingual Abstract)

In this paper, we propose a visual place recognition (VPR) technique that involves integrating textual information to overcome the challenges associated with repetitive patterns in indoor environments. As existing VPR algorithms mainly rely on visual features or geometric structures to recognize places, their performance degrades due to perceptual aliasing when they encounter repetitive patterns. Hence, we present a novel VPR algorithm that uses text, which is easily found in indoor environments, as a key feature to identify places. Since text recognition performance is critical in the proposed method, we adopt a sliding-window technique to extract consistent text for multiple query images to improve the text recognition accuracy and reduce the false positive rate. Moreover, the proposed method uses not only text but also the results of existing VPR algorithms, which allows it to overcome repetitive patterns while maintaining the performance of existing algorithms. Based on our experiments, the proposed algorithm achieves the lower false positive rates than existing VPR algorithms in indoor environments with repetitive patterns.